⟂ Comp2Close

Three tools behind one tab bar, built for a "Real Estate / Cross Assets" investment mandate — each one maps directly to a piece of the day-to-day workflow: sourcing relative value, drafting IC materials, and clearing due diligence.

Sheet	Tool	Maps to
1	Cross-Asset Comp Aggregator	Relative value across Real Assets, Private Debt, and Private Equity
2	IC Memo Drafter	"Translate quantitative insights into clear inputs for investment-committee materials"
3	DD Synthesizer	"Due diligence coordination... go/no-go documentation"

All three live in one HTML file (comp2close.html) served by one Flask app — switching sheets is instant, no reload, no separate logins.

---

Sheet 1 — Cross-Asset Comp Aggregator

Pulls comparable transaction and pricing data across Real Assets, Private Debt, and Private Equity into one dashboard — so relative value can be read across a platform, not within a single silo.

The question it answers: is capital priced richer in real estate cap rates right now, or in direct-lending yields, or in buyout multiples? That's the lens a cross-asset investment platform actually uses to allocate capital, and it's the reasoning this tool makes explicit and repeatable.

Why three asset classes, one dataset

Real estate cap rates, direct-lending spreads, and buyout multiples are expressed in incompatible units — you can't put a 4.5% cap rate and a 9.8x EV/EBITDA multiple on the same axis directly. So each data point is normalized to an index where 100 = the median reading for its own segment in that run. A bar above 100 means that particular reading is priced rich relative to its peer set; below 100 means it's priced cheap.

Output columns

asset_class · segment · geography · metric_type · metric_label · metric_value
data_type (Reported | Estimated) · as_of_date · trend
source_name · source_url · notes

• metric_type is a controlled field — each asset class has a fixed, enumerated list of allowed metric types, enforced both in the search prompt and in code:

  Asset class	Allowed metric types
  Real Assets	Prime Cap Rate · Secondary Cap Rate · Net Initial Yield · Reversionary Yield · Price per sqm · Price per sqft · Rental Growth Rate · Vacancy Rate · Total Return
  Private Debt	Unitranche Spread · Senior Term Loan Margin · Mezzanine Margin · All-in Yield · Original Issue Discount · Leverage Multiple (Debt/EBITDA) · Real Estate Debt Margin (by LTV)
  Private Equity	EV/EBITDA Multiple · EV/Revenue Multiple · Take-Private Premium · Entry Leverage (Debt/EBITDA) · Multiple Expansion/Compression

  A row claiming a Real Assets metric type on a Private Equity row (or any other cross-contamination) is dropped automatically — this is what makes the relative-value chart trustworthy rather than a blend of incompatible units.

• data_type is always disclosed: Reported or Estimated — never blended silently.

• source_url is the direct report/article; if unavailable, source_name names the publisher instead of a fabricated link.

---

Sheet 2 — IC Memo Drafter

Takes structured deal inputs — deal name, asset type, location, investment amount, rent roll/income summary, debt terms, business plan, target returns, known risks — and drafts a first-pass investment-committee memo with seven standard sections:

Executive Summary & Investment Thesis
Asset & Market Overview
Business Plan
Financial Summary
Returns & Sensitivities
Key Risks & Mitigants
Recommendation

Guardrails:

• Uses only what you type in — any missing field renders as "Not provided — to be confirmed" rather than a plausible-sounding invented figure.
• An optional checkbox lets it pull in whatever Sheet 1 last found as market context only (e.g. "entry cap rate is in line with current CBRE comps of ~4.3%") — it's explicitly told never to treat those comps as facts about your specific deal.
• Output is a clean printable document — Print / Save as PDF and Copy to clipboard buttons sit above the draft.
• Every memo is watermarked "Draft — figures as provided, not independently verified" so it's never mistaken for a finished, checked document.

---

Sheet 3 — DD Synthesizer

Paste (or upload as .txt/.md) text from three due-diligence document categories — Leases, Environmental & Technical Reports, Title & Legal Documents — plus optional deal context. The tool reads across all three and returns:

• An overall recommendation: Proceed / Proceed with Conditions / Do Not Proceed / Insufficient Information, with a short rationale
• A findings table: category, issue, severity (High/Medium/Low), the specific passage that triggered the flag, and a recommendation — sorted by severity
• Cross-document checks are explicitly prompted for — e.g. a lease break option that conflicts with the assumed hold period, or a title restriction that conflicts with the business plan implied by the lease terms

Guardrails:

• Findings are grounded strictly in the pasted text — nothing is invented for a category left blank; the tool says so rather than padding the list.
• Each finding cites the specific passage or clause that triggered it, so a reviewer can verify it against the source text directly.
• Accepts pasted content or .txt/.md/.pdf uploads (PDF text is extracted client-side via pdf.js). Scanned/image-only PDFs won't yield text — this is extraction, not OCR — and the tool flags that case explicitly rather than silently returning nothing.

---

Option A — try Sheet 1 right now in a Claude.ai chat (no setup)

Sheets 2 and 3 (IC Memo Drafter, DD Synthesizer) are interactive form-based tools best used in the dashboard itself — but Sheet 1's comp search works as a standalone chat prompt too. Turn Web search ON, then paste this:

You are a cross-asset research analyst at an alternative investment platform
(spanning Private Debt, Real Assets, and Private Equity) building a comp set
that shows relative value across those three asset classes.

Find current, publicly reported market pricing data for:
- Real Assets: prime office/logistics/residential/hotel cap rates in Europe
- Private Debt: direct lending unitranche pricing, real estate debt margins
- Private Equity: mid-market buyout and growth equity multiples in Europe

Search market reports and index providers: CBRE/JLL/Savills/Cushman & Wakefield
market reports and cap rate surveys, MSCI Real Assets, Green Street, RCA for
real assets; Deloitte Alternative Lender Deal Tracker, Cliffwater Direct Lending
Index, Preqin, PitchBook/LCD for private debt; Argos Index, Bain Global PE
Report, PitchBook European PE Breakdown for private equity. Run at least 4
searches per asset class.

For each data point, assign a metric_type from this controlled list (never mix
across asset classes):
- Real Assets: Prime Cap Rate, Secondary Cap Rate, Net Initial Yield, Reversionary
  Yield, Price per sqm, Price per sqft, Rental Growth Rate, Vacancy Rate, Total Return
- Private Debt: Unitranche Spread, Senior Term Loan Margin, Mezzanine Margin,
  All-in Yield, Original Issue Discount, Leverage Multiple (Debt/EBITDA), Real
  Estate Debt Margin (by LTV)
- Private Equity: EV/EBITDA Multiple, EV/Revenue Multiple, Take-Private Premium,
  Entry Leverage (Debt/EBITDA), Multiple Expansion/Compression

Only report figures traceable to a real source — mark each as Reported (hard
published figure) or Estimated (credible secondary estimate), name the source
either way, never invent a number.

Return a table: Asset Class | Segment | Metric Type | Geography | Value | Type
(Reported/Estimated) | As Of | Source. Then give me a one-paragraph relative-value
read: which asset class looks richly priced right now vs its own recent history,
and which looks cheap.

---

Option B — run the full toolkit (all 3 sheets)

pip install -r requirements.txt
export ANTHROPIC_API_KEY=sk-ant-...      # or copy .env.example to .env
python app.py   →   http://localhost:5000

One URL, three tabs: Cross-Asset Comps · IC Memo Drafter · DD Synthesizer. Switching sheets is instant (no reload) — this is the version worth demoing live rather than describing.

Sheet 1 also has a CLI, for scripted/scheduled comp pulls independent of the dashboard:

python agent.py
python agent.py --asset-classes "Real Assets" "Private Debt"
python agent.py --geography "UK and Ireland"
python agent.py --output ./results

Results save as cross_asset_comps_<timestamp>.json/.csv/.xlsx — the Excel file has one combined sheet plus one sheet per asset class, ready to drop into a Power BI or Tableau data model. Sheets 2 and 3 are dashboard-only (they're interactive, one-shot generation tools rather than scheduled pulls), so they don't have a separate CLI.

Sheets 2 and 3 also have CLIs now, for the same scripted/scheduled use case:

python memo_agent.py --deal-file deal.json --comps ./results/cross_asset_comps_20260709.json
python dd_agent.py --leases lease.pdf --env survey.pdf --title title.pdf

dd_agent.py reads .txt/.md/.pdf directly (PDF text extraction included — the browser tool does the equivalent client-side via pdf.js). See each script's --help for the full flag list.

---

Optional: export to Obsidian

obsidian_export.py converts any of the three CLIs' outputs into an Obsidian vault — one note per comp series / deal / DD review, with each re-run appending a dated history entry to the same note rather than creating a new file. See OBSIDIAN.md for setup, folder structure, and example Dataview queries.

---

Deploy to Render

render.yaml is included:

1. Push this repo to GitHub
2. Connect it at render.com → New Web Service
3. Paste ANTHROPIC_API_KEY (and optionally OPENAI_API_KEY) into the Render dashboard's environment variables
4. Render builds with pip install -r requirements.txt and serves with gunicorn app:app

All three sheets are live at the same deployed URL — nothing extra to configure.

---

Sources searched

Real Assets — CBRE / JLL / Savills / Cushman & Wakefield / Colliers market reports and cap rate surveys, MSCI Real Assets (UK Monthly Index), Green Street, RCA (Real Capital Analytics)

Private Debt — Deloitte Alternative Lender Deal Tracker, Cliffwater Direct Lending Index, Preqin, PitchBook/LCD, S&P LCD, Debtwire, direct lender published market commentary

Private Equity — Argos Index (European mid-market), Bain & Company Global Private Equity Report, PitchBook European PE Breakdown, EY/Deloitte annual PE reports

---

Further reading

See PROJECT_DOCUMENTATION.md for the methodology write-up: why relative value across asset classes matters, the normalization logic in detail, current limitations, and how this would extend into a production tool with licensed data feeds (Bloomberg, PitchBook API, Preqin API, MSCI).

---

A note on scope

Sheet 1 uses provider-side web search over public market reports and press — not licensed data feeds (Bloomberg, PitchBook, Preqin). Treat outputs as a directional snapshot for illustrating the relative-value framework, not as a substitute for a licensed data terminal in a live investment decision.

Sheet 2 drafts a genuine first pass only — it has no way to verify the figures you type in, and explicitly labels every output as an unverified draft. It is a starting point for an analyst to refine, not a finished IC document.

Sheet 3 reads pasted content or .txt/.md/.pdf uploads. PDF support is text extraction (pdf.js in the browser, pypdf in the CLI) — scanned/image-only PDFs still need OCR done separately before pasting in.